A long exposure of sky that initially looks empty reveals thousands of small bright patches. Many are galaxies, each with its own stars. Can we count those patches, multiply by the whole sky and obtain a final answer?
We do not know the exact number of galaxies. Estimates for the observable Universe depend on which objects are included and how unseen faint galaxies are accounted for. Scaling deep observations gives hundreds of billions; a well-known 2016 model estimate that includes numerous faint systems under specified assumptions gives about two trillion. These are different ways of estimating a population, not two exact censuses of which one must be wrong. Conselice and colleagues: methods, assumptions and counts.
First define what is being counted
The observable Universe is the region from which signals could have reached us over cosmic history. It is not a wall or a proven edge of all space. We cannot directly inventory galaxies beyond our observable region. A claim about their total number in the entire Universe would require assumptions about what we cannot see. American Museum of Natural History: the observable region and cosmic horizon.
We must also define the objects being counted. A large spiral galaxy is readily recognizable, but much fainter dwarf systems exist. Bright knots inside a distant galaxy can sometimes be mistakenly split into several sources, while nearby images of several galaxies can blend into one. Catalogs distinguish galaxies, foreground stars and artifacts; counting bright pixels is not enough. COSMOS2025: source detection and classification, Sections 3 and 6.3.
A deep field samples the sky; it is not the whole sky
The Hubble eXtreme Deep Field, XDF, released in 2012, contains about 5,500 galaxies within roughly one thirty-millionth of the whole sky. Accumulating light for a long time revealed objects that shorter exposures would miss. This is a historical deep-field example, not a claim that it remains the deepest of all today. ESA/Hubble: XDF area and galaxies.
Extrapolating a sample to the sky requires assumptions about how representative it is. Galaxies cluster: one field may contain more clusters, another more voids. Field-to-field scatter is called cosmic variance. Researchers therefore compare different fields and account for uncertainty, rather than declaring one photograph a precise miniature of the whole Universe. Conselice and colleagues: Appendix A.2.
Branch · With formulasOptional · you can skip this
What simple multiplication does—and does not—provide
Let n be the number of detected galaxies and f the fraction of sky occupied by the field. The simplest estimate assumes the same mean surface density:
N_(sky) ≈ (n)/(f).
Using rounded XDF quantities gives approximately 5500 × 30 000 000 = 165 000 000 000, or about 170 billion. Counts are dimensionless; f is also a dimensionless fraction, not an area in degrees.
This estimates how many sources of similar detectability might occupy the whole sky under the stated assumption. It omits missed faint objects, field differences, dust obscuration and other corrections. The result is a teaching extrapolation of ESA’s rounded example, not a new scientific census or a precise total of every galaxy.
Unseen faint objects do not vanish from the Universe
Imagine two observations of the same field. A less sensitive one detects only bright sources; a more sensitive one reveals some fainter patches. New galaxies need not have formed between the images: what changed is our ability to register their signal against noise.
Detection depends on received flux, wavelength range, image size and background. Even the same total flux is easier to find in a compact spot than when spread over a large area. Researchers test sample completeness: what fraction of objects with specified properties the method can find. COSMOS2025: completeness, contamination and depth comparisons.
Compare the panels. They show a deliberately constructed teaching population with all 20 objects known in advance; no mark is a real catalog galaxy.
Only the brightest appear
Detected: 6 of 20
14 fainter objects did not pass the hypothetical threshold. They did not cease to exist.
Fainter sources appear too
Detected: 20 of 20
In this deliberately simple example, the lower threshold finds all 20. Completeness is not guaranteed in a real survey.
All 20 schematic objects are specified by the author; this is not a photo, catalog or measurement of a telescope’s completeness. Galaxy positions match between panels. Selection changes the count, not physical birth or disappearance. Sensitivity levels are arbitrary, with no flux scale.
Beyond the sensitivity limit, researchers must estimate what the instrument did not individually detect. They use measured luminosity or stellar-mass distributions and extend a model to fainter or smaller systems. Such an extension is extrapolation. It is useful, but depends on the distribution’s form and the chosen lower limit. Conselice and colleagues: extrapolation caveats.
Branch · Limits of a familiar estimateOptional · you can skip this
Where two trillion came from
In their 2016 paper, Conselice and colleagues used galaxy stellar-mass distributions at different epochs and extended them to a lower limit of one million solar masses in stars. That is their counting threshold, not a definition under which every smaller system ceases to be a galaxy.
For the redshift range 0 to 8, they obtained a final estimate of approximately 2.0 trillion, with a quoted uncertainty interval of roughly 1.4 to 2.7 trillion. Redshift here specifies a range of cosmological epochs and distances; its detailed explanation is unnecessary for the central point.
This is not a catalog containing two trillion verified entries. The number depends substantially on the mass threshold and extending the model beyond the data. We also see distant systems at earlier epochs: a count along our view into the past is not a simultaneous photograph of the population throughout space “today.” Mergers and galaxy formation change their number over time. Original paper: Sections III.1 and IV.2–IV.3.
Better catalogs refine the sample, not deliver a final number
For example, the original COSMOS2025 paper describes a catalog of properties for more than 700,000 galaxies, built from JWST observations alongside other telescopes’ data. The NIRCam survey covers about 0.54 square degrees. Different analyses apply additional masks and selections, so their working area and counts do not always equal the survey’s full values. Shuntov and colleagues: the original catalog paper, 2025, COSMOS: public data release.
Such a catalog improves tests of faint-population models. But “the telescope found more” alone does not establish how the estimate for all galaxies changed. Comparable selection criteria, areas, wavelength ranges and completeness limits are needed.
Test your understanding: if a more sensitive image contains twice as many sources, has the Universe just doubled its galaxy population? No. Our accounting changed; physical evolution requires other evidence.
The answer to the opening question therefore includes scale and conditions: the observable Universe has an enormous galaxy population, estimated in hundreds of billions and, depending on faint-system models, at the trillion level. There is no exact final count, especially for the whole Universe. When a headline gives a new number, look for the survey area, sensitivity limit and treatment of unseen objects. To explore the numbers themselves, read about millions, billions and trillions.